We have completed the data reduction and merging of the electron diffraction intensities of crotoxin complex crystals to 3.5 _ resolution with over 350 patterns from 6 crystals. The Rmerge between the data and the sinc fitted curve along each lattice rod is around 27%. This value is similar to those obtained in other crystalline objects. However, a close examination of the individual curves shows their fit is poor. We then searched for the cause by examining the image data. Our current finding reveals that although the images diffract beyond 4 _, the unit cell motif within a single crystal appears varying based on the correlation map between a small reference image area and the entire crystal image. We are adopting the multivariate statistical and automatic classification analysis which has been used to process images of single particles. This computational method will evaluate the image data for their similarity and difference in a single crystal more readily than that of the conventional Fourier method. This approach may help us to uncover the cause of poor data merging, which is an essential step towards solving the high resolution structure of this thin crystal